2026

OpenAI Launches GPT-5.6 and New Model Family with Enhanced Security

Character.AI Launches Interactive Video Series Where Viewers Chat with Characters

Paris-based voice AI startup Gradium raises $100 million with Nvidia backing

Paris-based startup Gradium, which develops AI-powered speech synthesis and processing technologies, has received $100 million in seed fundi

01Google added a label for AI-generated ads in SearchJul 10, 2026·2 min 02OpenAI shuts down the Atlas browser less than a year into developmentJul 10, 2026·3 min 03Fidji Simo steps down as head of OpenAI’s AGI division due to illnessJul 10, 2026·3 min 04Character.AI launches AI vertical video series c.ai SeriesJul 10, 2026·2 min 05Meta opens access to Muse Spark 1.1 for AI coding through a new APIJul 10, 2026·2 min

Top themes

LLM·The Verge

Claude Reflect: Anthropic launches a Spotify Wrapped for ChatGPT

Anthropic has launched Claude Reflect, a dashboard for analyzing Claude usage that shows the main conversation topics, types of tasks perfor

01OpenAI released GPT-5.6 after government approval and introduced ChatGPT WorkJul 10, 2026·3 min 02A mistake in the law cost Estonia €28 million — now AI helps prevent such errorsJul 10, 2026·2 min 03UN AI Summit: can regulation keep pace with technologyJul 10, 2026·2 min

LLM·The Verge

Sunrun to offer customers payments for hosting AI computing nodes at home

U.S. energy company Sunrun is launching a pilot program under which owners of solar panels and batteries can receive payments for hosting AI

Australian Payments Plus sped up payments with ChatGPT Enterprise and Codex

The Australian payments company deployed ChatGPT Enterprise and Codex to automate payment processing while keeping humans in control of crit

OpenAI to train more than 1,600 U.S. K–12 educators in practical AI use

OpenAI Academy and the Walton Family Foundation will hold a series of in-person AI Skills Jam sessions for more than 1,600 K–12 educators, teaching them to use AI for lesson planning and school work.

OpenAI criticized the SWE-Bench Pro benchmark for inaccurate evaluation

OpenAI releases the GPT-5.6 family with three models and an ultra mode

OpenAI doubles reward for finding biological jailbreaks to $50,000

OpenAI unveils ChatGPT Work — an agent for multi-hour projects

Deutsche Telekom and OpenAI transform telecommunications with AI

AegisDx: AI framework for safe diagnostics outperformed GPT-5 in physicians’ evaluation

Harness engineering: moving LLM agent guarantees from prompts to code

How neural networks understand the values of different countries: a dataset of 500,000 examples

Structured LLM Pruning Without Accuracy Loss: Method Verified on Llama-3-8B and Vicuna

FMR Method Reduces RL Agent Alignment Errors by 98% in Imitation Learning

GRAPHEVAL: New Graph Framework Reveals That Self-Consistency Masks Reasoning Errors in LLMs

Self-Evolving LLM Agents: Compiling SOPs into Tools Reduces Latency by 42%

Researchers showed how pre-compiling standard SOP steps into verified tools reduces agent latency by 42% and error rates by 53% in real prod

01Graph neural network recognizes hand gestures in 48 ms: 99% accuracy for prosthetics and ARJul 10, 2026·3 min 02Qwen3 language models easily make educational tasks more complex, but struggle to simplify themJul 10, 2026·2 min 03NYIT developed an LLM assistant for an applied analysis course in MATLABJul 10, 2026·2 min 04Infinity-Parser2 surpassed DeepSeek-OCR-2 in document parsing and opened a dataset of 5 million examplesJul 10, 2026·3 min 05Hallucination Self-Play: a small language model surpassed large LLMs in hallucination detectionJul 10, 2026·2 min

Top themes

Gemini 2.5 Flash replaces human evaluators in voice agent testing

Researchers tested how reliably Gemini can evaluate voice AI agents instead of humans — and found that the model performs at 100 times lower

01Idiobionics: Scientists Warn of the Risks of Hacking Smart Bionic ProsthesesJul 10, 2026·2 min 02TACO: a method removes credit “contamination” in RL training of language modelsJul 10, 2026·2 min 03New MiniLM-based method more accurately detects queries outside chatbots' knowledge scopeJul 10, 2026·2 min

LLM·arXiv cs.AI

Scientists propose "alignment plausibility" as a safety standard for medical LLMs

A preprint on arXiv argues for a three-tier safety standard for language models in mental health, analogous to the regulation of physicians'

Debiasing NLP models creates new stereotypes for other demographic groups

An arXiv study found that preprocessing methods reduce bias for target groups but unintentionally amplify it for others — and standard bench

LLM·arXiv cs.AI

Review of 18 medical LLMs: narrow specialists diagnose more accurately than general-purpose models

Researchers compared 18 language models across a five-level scale of medical competencies — and found that specialized and general-purpose AI excel at different tasks.

EspanStereo: new dataset reveals LLM stereotypes in Spanish-speaking cultures

LSTM outperformed Random Forest and Naive Bayes in tweet sentiment analysis with a ROC-AUC of 0.92

Context Graph: researchers taught enterprise AI agents to work without queries

LiST: a new method simultaneously delivers accuracy, robustness, and calibration in neural networks

Hol-PCFG cut syntactic parsing parameters by 99.94% and took the lead in six languages

MetaNCA teaches neural networks to self-organize without backpropagation

Jet-Long extends language model context windows to 128K without fine-tuning

Eternis-Forecaster 8B: activation probes are more accurate than reasoning chains and save up to 47% of tokens

Researchers showed that probes of intermediate LLM activations reveal actual forecasts more accurately than chain-of-thought and can save 30–47% of tokens without losing accuracy.